Changing the size of allowed gaps did however have an impact on the results, but only when incrementing at smaller values. Varying this parameter from 0 to 1 varied the total number of clustered genes between 1 and 3 percent within the various species and this effect diminished as the parameter was increased. An effect of less than a tenth of a percent was obtained for all species by varying the gap length from 14 to 15, so a gap length of 15 is the value we chose to implement as a threshold across all species. Nested tandem arrays of a different family were counted as only one gap space in this analysis. We evaluated the total amount of clustering as a function of the stringency of the expectation threshold required for selecting a chain as significant by performing multiple chromosome walks and applying various expectation thresholds in both the real and randomly permuted genomes. Empirically we found that requiring e,0.01 practically eliminated the detection of gene clusters among randomized genomes in all the annotation systems tested across all species. This provided a quite conservative threshold for paracluster detection, with a low probability of false positive clusters, and resulting in a minimum estimation of genome wide Perifosine paraclustering metrics. Furthermore, at the e,0.01 threshold, essentially all clusters greater than two genes were identified. When evaluating higher thresholds the tendency in cluster identification was to include additional clusters containing only two genes having increasing space between them. Empirically we found that those datasets which involved whole gene annotation leveled off in total clustering as the threshold was increased to e,0.1 and greater. Any increase in the total genome wide clustering metrics above this threshold was almost all due to the domain specific annotation datasets, namely InterPro and SCOP. Because this method is dependent on the quality and extensiveness of annotation data for each of the studied genomes, we evaluated the impact of genes in a given species having no annotations in any of the datasets, consequently creating dark MK-0683 HDAC inhibitor regions in the genome according to the methodology. Actually, for each species we obtained 100 percent annotation coverage simply due to the inclusion of the Ensembl family annotations which involve all genes, but we investigated whether eliminating Ensembl family annotations from the analysis otherwise revealed large annotation gaps. Doing this, we found that some genomes had less coverage than others; for example the chicken genome and the fly genome, having the smallest coverage, had only around 86% of mapped genes in total coverage whereas humans had 93%. But we were able to conclude based on specific characteristics of these regions that their contribution to total paraclustering would be small and their impact would be inconsequential to the total reported metrics.